geeta: gold standard annotated data, analysis and its...
TRANSCRIPT
. . . . . .
.
......
Geeta: Gold Standard Annotated Data,Analysis and its Application
Preeti ShuklaAmba KulkarniDevanand Shukl
Department of Sanskrit Studies,University of Hyderabad,
Hyderabad
20/12/2013
P. Shukla, A. Kulkarni, D. Shukl (UoHyd) Geeta 1 / 45 20/12/2013 1 / 45
. . . . . .
Outline
...1 Gold standard data and its advantages
...2 Levels of Tagging
...3 Methodology
...4 Quantitative Analysis of BhG
...5 Utility and Conclusion
P. Shukla, A. Kulkarni, D. Shukl (UoHyd) Geeta 2 / 45 20/12/2013 2 / 45
. . . . . .
Gold standard data and its advantages
Gold standard data
When a corpus is manually annotated at various levels, such as Treebank annotation, Discourse level annotation, etc., then it serves as agold standard data for evaluation and comparison of various tools.This has the following advantages –
P. Shukla, A. Kulkarni, D. Shukl (UoHyd) Geeta 3 / 45 20/12/2013 3 / 45
. . . . . .
Gold standard data and its advantages
Advantages
...1 One can use the gold standard annotated data as an input forthe evaluation of various modules. This avoids the cascadingeffects and one can measure the absolute performance of varioustools.
...2 To use data-driven models to tune the machine for betterperformance in a chosen domain.For eg., Penn Tree bank
P. Shukla, A. Kulkarni, D. Shukl (UoHyd) Geeta 4 / 45 20/12/2013 4 / 45
. . . . . .
Gold standard data and its advantages
Advantages
...1 One can use the gold standard annotated data as an input forthe evaluation of various modules. This avoids the cascadingeffects and one can measure the absolute performance of varioustools.
...2 To use data-driven models to tune the machine for betterperformance in a chosen domain.For eg., Penn Tree bank
P. Shukla, A. Kulkarni, D. Shukl (UoHyd) Geeta 4 / 45 20/12/2013 4 / 45
. . . . . .
Gold standard data and its advantages
Śrīmad Bhagvad Gītā
We chose Śrīmad Bhagvad Gītā (BhG for short) consisting of 18chapters and 700 verses (ślokas) for developing a gold standard datafor Sanskrit text mainly because:
It is an important text which summarizes the Upanishadicteachings and is commented upon and interpreted by variousschools of Indian philosophies.Thus annotators in doubt can always refer to thesecommentaries for correct annotation.
P. Shukla, A. Kulkarni, D. Shukl (UoHyd) Geeta 5 / 45 20/12/2013 5 / 45
. . . . . .
Gold standard data and its advantages
This scripture being coherent and complete in itself, can be usedfor higher level analysis such as discourse analysis, topicidentification, anaphora resolution, and so on.
P. Shukla, A. Kulkarni, D. Shukl (UoHyd) Geeta 6 / 45 20/12/2013 6 / 45
. . . . . .
Gold standard data and its advantages
Further, this also being part of the Mahābhārata (section 25 to42 of the Bhiṣma parva), later on if necessary, it can be used asan initial training data for boot-strapping for automaticannotation of complete critical edition of Mahābhārata (witharound hundred thousand verses).
P. Shukla, A. Kulkarni, D. Shukl (UoHyd) Geeta 7 / 45 20/12/2013 7 / 45
. . . . . .
Levels of Tagging
Sanskrit requires the following levels of tagging viz.,...1 Annotation of Sandhi...2 Tagging of Compounds...3 Morphological Analysis...4 Tagging of Sentential relations...5 Marking the Prose order
P. Shukla, A. Kulkarni, D. Shukl (UoHyd) Geeta 8 / 45 20/12/2013 8 / 45
. . . . . .
Levels of Tagging
Sanskrit requires the following levels of tagging viz.,...1 Annotation of Sandhi...2 Tagging of Compounds...3 Morphological Analysis...4 Tagging of Sentential relations...5 Marking the Prose order
P. Shukla, A. Kulkarni, D. Shukl (UoHyd) Geeta 8 / 45 20/12/2013 8 / 45
. . . . . .
Levels of Tagging
Sanskrit requires the following levels of tagging viz.,...1 Annotation of Sandhi...2 Tagging of Compounds...3 Morphological Analysis...4 Tagging of Sentential relations...5 Marking the Prose order
P. Shukla, A. Kulkarni, D. Shukl (UoHyd) Geeta 8 / 45 20/12/2013 8 / 45
. . . . . .
Levels of Tagging
Sanskrit requires the following levels of tagging viz.,...1 Annotation of Sandhi...2 Tagging of Compounds...3 Morphological Analysis...4 Tagging of Sentential relations...5 Marking the Prose order
P. Shukla, A. Kulkarni, D. Shukl (UoHyd) Geeta 8 / 45 20/12/2013 8 / 45
. . . . . .
Levels of Tagging
Sanskrit requires the following levels of tagging viz.,...1 Annotation of Sandhi...2 Tagging of Compounds...3 Morphological Analysis...4 Tagging of Sentential relations...5 Marking the Prose order
P. Shukla, A. Kulkarni, D. Shukl (UoHyd) Geeta 8 / 45 20/12/2013 8 / 45
. . . . . .
Levels of Tagging
Annotation of Sandhi
The first level of tagging needed for Sanskrit is the marking of wordboundaries undoing the sandhi (euphonic changes). We split twotypes of sandhis –
...1 sandhi between two words which is indicated by a ‘+’ sign
...2 sandhi between the components of a compound is indicated by a‘-’ sign
P. Shukla, A. Kulkarni, D. Shukl (UoHyd) Geeta 9 / 45 20/12/2013 9 / 45
. . . . . .
Levels of Tagging
Annotation of Sandhi
The first level of tagging needed for Sanskrit is the marking of wordboundaries undoing the sandhi (euphonic changes). We split twotypes of sandhis –
...1 sandhi between two words which is indicated by a ‘+’ sign
...2 sandhi between the components of a compound is indicated by a‘-’ sign
P. Shukla, A. Kulkarni, D. Shukl (UoHyd) Geeta 9 / 45 20/12/2013 9 / 45
. . . . . .
Levels of Tagging
Example
.BhG 2.40 –..
......nehābhikramanāśo’asti pratyavāyo na vidyate|svalpamapyasya dharmasya trāyate mahato bhayāt||
Eng: In this path there is no loss of effort, nor is there any adverseresult. Even a little practice of this discipline protects one from greatfear (of birth and death).
P. Shukla, A. Kulkarni, D. Shukl (UoHyd) Geeta 10 / 45 20/12/2013 10 / 45
. . . . . .
Levels of Tagging
Verse tokenized as
.
......na+iha+abhikrama-nāśaḥ+asti pratyavāyaḥ+na vidyate|svalpam+api+asya dharmasya trāyate mahataḥ+bhayāt||
P. Shukla, A. Kulkarni, D. Shukl (UoHyd) Geeta 11 / 45 20/12/2013 11 / 45
. . . . . .
Levels of Tagging
Tagging of Compounds
Sanskrit compounds are broadly classified as follows:...1 endo-centric / adverbial (tatpuruṣaḥ / avyayībhāvaḥ)...2 exo-centric (bahuvrīhiḥ)...3 copulative (karmadhārayaḥ)...4 conjunctive (dvandvaḥ)
These compounds with an exception of dvandva (conjunctive) andbahuvrīhi (exocentric) are binary.
P. Shukla, A. Kulkarni, D. Shukl (UoHyd) Geeta 12 / 45 20/12/2013 12 / 45
. . . . . .
Levels of Tagging
Tagging of Compounds
Sanskrit compounds are broadly classified as follows:...1 endo-centric / adverbial (tatpuruṣaḥ / avyayībhāvaḥ)...2 exo-centric (bahuvrīhiḥ)...3 copulative (karmadhārayaḥ)...4 conjunctive (dvandvaḥ)
These compounds with an exception of dvandva (conjunctive) andbahuvrīhi (exocentric) are binary.
P. Shukla, A. Kulkarni, D. Shukl (UoHyd) Geeta 12 / 45 20/12/2013 12 / 45
. . . . . .
Levels of Tagging
Tagging of Compounds
Sanskrit compounds are broadly classified as follows:...1 endo-centric / adverbial (tatpuruṣaḥ / avyayībhāvaḥ)...2 exo-centric (bahuvrīhiḥ)...3 copulative (karmadhārayaḥ)...4 conjunctive (dvandvaḥ)
These compounds with an exception of dvandva (conjunctive) andbahuvrīhi (exocentric) are binary.
P. Shukla, A. Kulkarni, D. Shukl (UoHyd) Geeta 12 / 45 20/12/2013 12 / 45
. . . . . .
Levels of Tagging
Tagging of Compounds
Sanskrit compounds are broadly classified as follows:...1 endo-centric / adverbial (tatpuruṣaḥ / avyayībhāvaḥ)...2 exo-centric (bahuvrīhiḥ)...3 copulative (karmadhārayaḥ)...4 conjunctive (dvandvaḥ)
These compounds with an exception of dvandva (conjunctive) andbahuvrīhi (exocentric) are binary.
P. Shukla, A. Kulkarni, D. Shukl (UoHyd) Geeta 12 / 45 20/12/2013 12 / 45
. . . . . .
Levels of Tagging
The meaning of a compound is decided by the way the componentsare combined together.A compound with 3 components a-b-c may be combined in twodifferent ways viz.,
<<a-b>-c><a-<b-c>>
P. Shukla, A. Kulkarni, D. Shukl (UoHyd) Geeta 13 / 45 20/12/2013 13 / 45
. . . . . .
Levels of Tagging
Example
The compound word dehāntaraprāptiḥ in BhG 2.13 is combined as<<a-b>-c>anyaḥ dehaḥ – dehāntaramGloss: transference_of_the_bodydehāntarasya prāptiḥ – dehāntaraprāptiḥGloss: attainment_of_transference_of_the_body
P. Shukla, A. Kulkarni, D. Shukl (UoHyd) Geeta 14 / 45 20/12/2013 14 / 45
. . . . . .
Levels of Tagging
Example
The compound word dehāntaraprāptiḥ in BhG 2.13 is combined as<<a-b>-c>anyaḥ dehaḥ – dehāntaramGloss: transference_of_the_bodydehāntarasya prāptiḥ – dehāntaraprāptiḥGloss: attainment_of_transference_of_the_body
P. Shukla, A. Kulkarni, D. Shukl (UoHyd) Geeta 14 / 45 20/12/2013 14 / 45
. . . . . .
Levels of Tagging
Various stages
Kumar et al., (2010) describe various stages involved in the analysisof a compound which also form the natural modules of a compoundprocessor –
...1 Segmentation (samāsapadacchedadh)
...2 Constituency Parsing (samāsapadānvayaḥ)
...3 Compound Type Identification (samastapadaparicāyakaḥ)
...4 Paraphrasing (vigraha-vākyam)A hierarchical tagset of 55 tags has been designed to tag the Sanskritcompounds.
P. Shukla, A. Kulkarni, D. Shukl (UoHyd) Geeta 15 / 45 20/12/2013 15 / 45
. . . . . .
Levels of Tagging
Various stages
Kumar et al., (2010) describe various stages involved in the analysisof a compound which also form the natural modules of a compoundprocessor –
...1 Segmentation (samāsapadacchedadh)
...2 Constituency Parsing (samāsapadānvayaḥ)
...3 Compound Type Identification (samastapadaparicāyakaḥ)
...4 Paraphrasing (vigraha-vākyam)A hierarchical tagset of 55 tags has been designed to tag the Sanskritcompounds.
P. Shukla, A. Kulkarni, D. Shukl (UoHyd) Geeta 15 / 45 20/12/2013 15 / 45
. . . . . .
Levels of Tagging
Various stages
Kumar et al., (2010) describe various stages involved in the analysisof a compound which also form the natural modules of a compoundprocessor –
...1 Segmentation (samāsapadacchedadh)
...2 Constituency Parsing (samāsapadānvayaḥ)
...3 Compound Type Identification (samastapadaparicāyakaḥ)
...4 Paraphrasing (vigraha-vākyam)A hierarchical tagset of 55 tags has been designed to tag the Sanskritcompounds.
P. Shukla, A. Kulkarni, D. Shukl (UoHyd) Geeta 15 / 45 20/12/2013 15 / 45
. . . . . .
Levels of Tagging
Various stages
Kumar et al., (2010) describe various stages involved in the analysisof a compound which also form the natural modules of a compoundprocessor –
...1 Segmentation (samāsapadacchedadh)
...2 Constituency Parsing (samāsapadānvayaḥ)
...3 Compound Type Identification (samastapadaparicāyakaḥ)
...4 Paraphrasing (vigraha-vākyam)A hierarchical tagset of 55 tags has been designed to tag the Sanskritcompounds.
P. Shukla, A. Kulkarni, D. Shukl (UoHyd) Geeta 15 / 45 20/12/2013 15 / 45
. . . . . .
Levels of Tagging
Morphological Analysis
At this stage tagging of both the inflectional as well as derivationalinformation is needed. e.g.,The morph of the verb ‘asti’ (to be) is.V...as2{active;laṭ;1st;sg;parasmaipadī;asaz;adādiḥ}
P. Shukla, A. Kulkarni, D. Shukl (UoHyd) Geeta 16 / 45 20/12/2013 16 / 45
. . . . . .
Levels of Tagging
Tagging of Sentential relations
In order to interpret the meaning of a sentence, various relationsamong words are necessary.Of around 90 relations classified by Ramakrishnamacharyulu (2009),only around 35 relations based on syntactico-semantic informationavailable in a sentence are considered for automatic tagging.
P. Shukla, A. Kulkarni, D. Shukl (UoHyd) Geeta 17 / 45 20/12/2013 17 / 45
. . . . . .
Levels of Tagging
Marking the Prose order
For understanding a Sanskrit text in verse style, two differentmethods have been followed in Indian education system viz.,
Daṇḍānvaya(also known as anvayamukhī)
The teacher arranges all the words in prose order for easyunderstanding of a verse.This approach assumes that if a user is given the ‘default proseorder’ of the sentence, he ‘understands’ its meanings.
P. Shukla, A. Kulkarni, D. Shukl (UoHyd) Geeta 18 / 45 20/12/2013 18 / 45
. . . . . .
Levels of Tagging
Example
.BhG 4.8 –..
......paritrāṇāya sādhūnāṃ vināshāya ca duṣkṛtām|dharmasaṃsthāpanārthāya sambhavāmi yuge yuge||
Anvaya: aham sādhūnāṃ paritrāṇāya duṣkṛtām vināshāyadharmasaṃsthāpanārthāya ca yuge yuge sambhavāmi
Eng: I appear from time to time for protecting the good, fortransforming the evil-minded, and for establishing world order(Dharma).
P. Shukla, A. Kulkarni, D. Shukl (UoHyd) Geeta 19 / 45 20/12/2013 19 / 45
. . . . . .
Levels of Tagging
Khaṇḍānvaya(also known as kathambhūtinī)
The teacher gives the basic skeleton of a sentence and fills inother details by asking questions which are centered around theheads seeking their various modifiers.This approach is close to parsing a sentence showing variousdependency relations.
P. Shukla, A. Kulkarni, D. Shukl (UoHyd) Geeta 20 / 45 20/12/2013 20 / 45
. . . . . .
Levels of Tagging
Canonical Form
The default word order or the ‘canonical form’ is governed roughly bythe following verse:.samāsacakram kā.verse 10..
......viśeṣaṇam puraskṛtya viśeṣyam tad-lakṣaṇam|kartṛ-karma-kriyā-yuktam etad anvaya-lakṣaṇam||
gloss: Starting with the adjectives, targeting the headword, in theorder of kartṛ-karma-kriyā (subject-object-verb) gives an anvaya.
P. Shukla, A. Kulkarni, D. Shukl (UoHyd) Geeta 21 / 45 20/12/2013 21 / 45
. . . . . .
Levels of Tagging
Example – BhG 2.40
<l><seg type=“pāda”><euphonic-word no=“1”> nehābhikramanāśo’sti<word no=“1” prose word order_no=“3”> na<mo_anal> na{ind} </mo_anal><syntactic_rel> mod 4 </syntactic_rel></word><word no=“2” prose word order_no = “1”> iha<mo_anal> iha{ind} </mo_anal><syntactic_rel> loc 4 </syntactic_rel></word>
P. Shukla, A. Kulkarni, D. Shukl (UoHyd) Geeta 22 / 45 20/12/2013 22 / 45
. . . . . .
Levels of Tagging
<word no=“3” prose word order_no=“2”> abhikramanāśaḥ<compound label=“T6”><component no=“1”> abhikrama </component><component no=“2”> nāśaḥ<mo_anal> nāśa{masc}{nom;sg} </mo_anal></component></compound><syntactic_rel> subj 4 </syntactic_rel></word><word no=“4” prose word order_no=“4” > asti<mo_anal>as2{active;laṭ;1st;sg;parasmaipadī;asaz;adādiḥ} </mo_anal></word></euphonic-word></seg></l>
P. Shukla, A. Kulkarni, D. Shukl (UoHyd) Geeta 23 / 45 20/12/2013 23 / 45
. . . . . .
Methodology
The process for semi-automatic tagging of BhG is as follows:The verse form is converted into prose form.
P. Shukla, A. Kulkarni, D. Shukl (UoHyd) Geeta 24 / 45 20/12/2013 24 / 45
. . . . . .
Methodology
Initially the sandhi and compound in the verse are segmentedmanually, following the guidelines developed by the SHMTconsortium 1. Then each compound is tagged for its type, alongwith the complete constituency mark-up.
1This is the Consortium of 7 institutes, for ‘Development ofSanskrit-Hindi Machine Translation System (sampark)’ funded by DIT,Govt. of India
P. Shukla, A. Kulkarni, D. Shukl (UoHyd) Geeta 25 / 45 20/12/2013 25 / 45
. . . . . .
Methodology
The segmented words are run in the anusāraka interface 2 forobtaining the multiple morph analysis. The output generated asan xml file, is then manually pruned for choosing the correctmorph analysis in the context.
2http://sanskrit.uohyd.ernet.in/sclP. Shukla, A. Kulkarni, D. Shukl (UoHyd) Geeta 26 / 45 20/12/2013 26 / 45
. . . . . .
Methodology
The synactico-semantic relations are tagged manually, followingthe guidelines develped by the SHMT consortium 3.
3http://sanskrit.uohyd.ernet.in/scl/Corpus/TaggingGuidelines/kaaraka-tagging-guidelines
P. Shukla, A. Kulkarni, D. Shukl (UoHyd) Geeta 27 / 45 20/12/2013 27 / 45
. . . . . .
Methodology
Hindi and English glosses for each word are given manually. Forthis we followed Geeta Press 4.
4Śrīmad Bhagvad Gītā, Geeta Press, Gorakhpur, India, reprint 2007P. Shukla, A. Kulkarni, D. Shukl (UoHyd) Geeta 28 / 45 20/12/2013 28 / 45
. . . . . .
Quantitative Analysis of BhG
Compound Distribution
Compound-type FreqEndocentric 994Exocentric 390Copulative 163Conjunctive 144
.. Refer DSP
P. Shukla, A. Kulkarni, D. Shukl (UoHyd) Geeta 29 / 45 20/12/2013 29 / 45
. . . . . .
Quantitative Analysis of BhG
Morphological statistics:
multiple multiplemorph count words morph count words
1 5280 8 222 1570 9 283 1082 10 154 349 11 105 292 12 76 99 13 37 121 14 6
Total words 8884Average 1.90
P. Shukla, A. Kulkarni, D. Shukl (UoHyd) Geeta 30 / 45 20/12/2013 30 / 45
. . . . . .
Quantitative Analysis of BhG
Case Statistics in Geeta
case singular dual pluralnom. 2463 31 613acc. 1349 20 251instr. 266 0 94dat. 57 0 5abl. 116 0 12gen. 335 24 179loc. 273 3 92voc. 251 1 0
P. Shukla, A. Kulkarni, D. Shukl (UoHyd) Geeta 31 / 45 20/12/2013 31 / 45
. . . . . .
Quantitative Analysis of BhG
Case-Number distribution in context
case singular dual pluralnom. 864 29 501acc. 733 20 103instr. 78 0 48dat. 29 0 5abl. 53 0 12gen. 222 24 98loc. 155 3 48voc. 80 1 0total 2014 74 795
.. Refer DSP
P. Shukla, A. Kulkarni, D. Shukl (UoHyd) Geeta 32 / 45 20/12/2013 32 / 45
. . . . . .
Quantitative Analysis of BhG
syntactico-semantic relations in Geetarelation freq relation freqviśeṣaṇa (adjective) 1277 kartā (subject) 1256conjunctive 1155 karma (object) 924predicative adj 401 adhikaraṇa (locative) 358ṣaṣṭhi (genitive) 357 negation 242emphatic 275 sambodhana (vocative) 237precedence 194 adverb 194karaṇa (instrument) 130 karma- 113
sāmānidhikaraṇahetu (causal) 96 co-relative 82
sarvanāma (pronouns)apādāna (source) 75 vākya-karma 64prayojana (purpose) 52 simultaneity 50sampradāna (dative) 15 disjunction 10
P. Shukla, A. Kulkarni, D. Shukl (UoHyd) Geeta 33 / 45 20/12/2013 33 / 45
. . . . . .
Quantitative Analysis of BhG
Tense-Mood distribution in GeetaSanskrit has 10 lakāras which represent Tense-Modality.
lakāra freqlaṭ (Present) 355loṭ (Imperative) 72lṛṭ (Second future) 42vidhiliṅ (Potential) 40laṅ (Imperfect) 26liṭ (Perfect) 22luṭ (First future) 16luṅ (Aorist) 9āśīrliṅ (Optative) 6lṛṅ (Conditional) 1
.. Refer DSP
P. Shukla, A. Kulkarni, D. Shukl (UoHyd) Geeta 34 / 45 20/12/2013 34 / 45
. . . . . .
Utility and Conclusion
Utility
There are three important usages of this gold data viz.,...1 for NLP applications...2 as linguistic inputs for developing a domain specific primer for
learning BhG...3 with the suitable interface, a self-learning / reading tool for BhG.
P. Shukla, A. Kulkarni, D. Shukl (UoHyd) Geeta 35 / 45 20/12/2013 35 / 45
. . . . . .
Utility and Conclusion
Utility
There are three important usages of this gold data viz.,...1 for NLP applications...2 as linguistic inputs for developing a domain specific primer for
learning BhG...3 with the suitable interface, a self-learning / reading tool for BhG.
P. Shukla, A. Kulkarni, D. Shukl (UoHyd) Geeta 35 / 45 20/12/2013 35 / 45
. . . . . .
Utility and Conclusion
Utility
There are three important usages of this gold data viz.,...1 for NLP applications...2 as linguistic inputs for developing a domain specific primer for
learning BhG...3 with the suitable interface, a self-learning / reading tool for BhG.
P. Shukla, A. Kulkarni, D. Shukl (UoHyd) Geeta 35 / 45 20/12/2013 35 / 45
. . . . . .
Utility and Conclusion
NLP applications
serves as a gold standard for evaluation of various Sanskrit tools.for developers of NLP tools.in prioritizing the solutions in the case of ambiguities.
P. Shukla, A. Kulkarni, D. Shukl (UoHyd) Geeta 36 / 45 20/12/2013 36 / 45
. . . . . .
Utility and Conclusion
NLP applications
serves as a gold standard for evaluation of various Sanskrit tools.for developers of NLP tools.in prioritizing the solutions in the case of ambiguities.
P. Shukla, A. Kulkarni, D. Shukl (UoHyd) Geeta 36 / 45 20/12/2013 36 / 45
. . . . . .
Utility and Conclusion
NLP applications
serves as a gold standard for evaluation of various Sanskrit tools.for developers of NLP tools.in prioritizing the solutions in the case of ambiguities.
P. Shukla, A. Kulkarni, D. Shukl (UoHyd) Geeta 36 / 45 20/12/2013 36 / 45
. . . . . .
Utility and Conclusion
NLP applications
serves as a gold standard for evaluation of various Sanskrit tools.for developers of NLP tools.in prioritizing the solutions in the case of ambiguities.
P. Shukla, A. Kulkarni, D. Shukl (UoHyd) Geeta 36 / 45 20/12/2013 36 / 45
. . . . . .
Utility and Conclusion
Domain Specific Primer
The quantitative analysis can help a teacher to decide whichaspect of Sanskrit Grammar is more relevant for the study ofBhG as follows:
...1 postponing the teaching of dual forms to a later stage.
...2 deciding how many and which paradigms of noun declension toconcentrate on first.
...3 using the analysed data of Tense-Modality to decide whichlakāras to teach and which conjugation classes to concentrateon first.
...4 deciding on which type of compound to teach first based on thecompound distribution table.
.DSP .. Go to Case .. Go to lakara .. Go to CD
P. Shukla, A. Kulkarni, D. Shukl (UoHyd) Geeta 37 / 45 20/12/2013 37 / 45
. . . . . .
Utility and Conclusion
Domain Specific Primer
The quantitative analysis can help a teacher to decide whichaspect of Sanskrit Grammar is more relevant for the study ofBhG as follows:
...1 postponing the teaching of dual forms to a later stage.
...2 deciding how many and which paradigms of noun declension toconcentrate on first.
...3 using the analysed data of Tense-Modality to decide whichlakāras to teach and which conjugation classes to concentrateon first.
...4 deciding on which type of compound to teach first based on thecompound distribution table.
.DSP .. Go to Case .. Go to lakara .. Go to CD
P. Shukla, A. Kulkarni, D. Shukl (UoHyd) Geeta 37 / 45 20/12/2013 37 / 45
. . . . . .
Utility and Conclusion
Self Learning cum Reading Tool
With the help of a suitable interface such as that of anusāraka, aninterested reader can have complete analysis of BhG at various levels.The interface provides –
P. Shukla, A. Kulkarni, D. Shukl (UoHyd) Geeta 38 / 45 20/12/2013 38 / 45
. . . . . .
Utility and Conclusion
User controlled access to various levels of analysis (Figs. 1,2,3).The graphs showing the constituency information and the kārakarelations are generated automatically from the manually taggeddata.
Figure: 1. compound analysis of BhG 2.40
P. Shukla, A. Kulkarni, D. Shukl (UoHyd) Geeta 39 / 45 20/12/2013 39 / 45
. . . . . .
Utility and Conclusion
Figure: 2. kāraka analysis of BhG 2.40
P. Shukla, A. Kulkarni, D. Shukl (UoHyd) Geeta 40 / 45 20/12/2013 40 / 45
. . . . . .
Utility and Conclusion
Figure: 3. morph analysis of BhG 2.40
P. Shukla, A. Kulkarni, D. Shukl (UoHyd) Geeta 41 / 45 20/12/2013 41 / 45
. . . . . .
Utility and Conclusion
Link to various dictionaries for meanings of the head words.Graphical display of phrase structure analysis of compounds (Fig4).
Figure: 4. Compound analysis of a word from BhG 12.17
P. Shukla, A. Kulkarni, D. Shukl (UoHyd) Geeta 42 / 45 20/12/2013 42 / 45
. . . . . .
Utility and Conclusion
Graphical display of sentential analysis (Fig 5).
Figure: 5. Dependency graph of BhG 2.40
P. Shukla, A. Kulkarni, D. Shukl (UoHyd) Geeta 43 / 45 20/12/2013 43 / 45
. . . . . .
Utility and Conclusion
This provides the user a digitized learning and understandingenvironment and forms a basis for the theoretical linguistsand grammarians to test their theories as well.
P. Shukla, A. Kulkarni, D. Shukl (UoHyd) Geeta 44 / 45 20/12/2013 44 / 45
. . . . . .
Utility and Conclusion
धन्यवादः!!!
P. Shukla, A. Kulkarni, D. Shukl (UoHyd) Geeta 45 / 45 20/12/2013 45 / 45